About this project

vtmate is a powerful, self-contained terminal-based voice AI toolkit designed for low-latency conversational interactions with local or cloud-based LLMs. It integrates speech-to-text (Whisper) and text-to-speech (Kokoro, Supertonic) engines directly, requiring no external installations for core functionality. The system supports 41 languages, voice cloning, and extensive agent customization via reusable system prompts. Key features include: - **Conversation Modes**: Supports Live (continuous voice with interruption) and Push-to-Talk (PTT) modes. Users can interrupt responses, reset sessions, or undo previous turns. - **Debate Mode**: Enables automated debates between two AI agents, with user participation possible. - **Read Mode**: Converts text files or STDIN input into speech with phrase-by-phrase navigation and pause/resume controls. - **Daemon Mode**: Runs in the background, allowing global shortcuts for voice chatting, asking questions about selected text, reading selections aloud, or dictating text into any application. - **LLM Integration**: Compatible with local servers (Ollama, llama.cpp) and hosted providers (OpenAI, Anthropic, Google, etc.). - **Session Export**: Saves conversations as audio/text files or interactive HTML players for playback. Installation is streamlined via a single curl command, and the tool is portable across macOS, Linux, and Windows. vtmate aims to replace paid voice AI tools by providing a comprehensive, offline-capable, and highly customizable voice interface for terminal environments.